Author : Belsam Jeba Ananth, K.G. Revathi, D. Poornima,
A NOVEL VIDEO COMPRESSION THEN RESTORATION ARTIFACT REDUCTION METHOD BASED ON OPTICAL FLOW CONSISTENCY TO IMPROVE THE QUALITY OF THE VIDEO
European Journal of Molecular & Clinical Medicine,
2020, Volume 7, Issue 3, Pages 5463-5474
In the recent days, the multimedia communication is created enormous impact in transmission and reception of data. In the multimedia communication, the video transmission, reception and storage are playing significant role. The requirement to store the video into the drive requires huge space and memory. The video compression methodology is segmented into two classes; first is lossy compression and next is lossless compression. In the lossy compression, the video frames are compressed high comparatively than lossless compression but after than compression, the video quality becomes poor due to the occurrence of artifacts. In this research, an innovative methodology is implemented to compress the video frame by applying the artifact reduction technique. The video frame data compression techniques could be significantly applied for reducing the huge size of video frame content, but they also produce objectionable visual artifacts due to making the lossy compression. The proposed methodology is used to enhance the quality of the video after compression by applying Deep Recursive Ensemble Particle Filter (DREPF) technique to remove artifacts. Particularly, the video quality improvement after compression is designed as a Recursive Ensemble Particle Filtering (REPF) process and the decoded video frames could be enhanced from the proposed DREPF. In the proposed methodology, Deep Convolutional Neural Network (DCNN) is implemented to calculate the equivalent positions in the Recursive Ensemble Particle Filter (REPF) and incorporated mutually in the deep Recursive Ensemble Particle Filtering. More significantly, the preceding information is estimated by integrating the time and temporal system for better enhancement of video frames. The proposed methodology obtains the benefits of the model-based system and learning-based system, by incorporating the recursive nature of the Recursive Ensemble Particle model and dominant illustration capability of DCNN. The results of experiments show that the proposed methodology gives better results than existing systems.